Structural insights into the degradation of Mcl-1 induced by BH3 domains

Noxa Causes Mcl-1 Degradation, but Bim Causes Mcl-1 Stabilization.

As previously shown (15), overexpression of human Noxa triggered the degradation of endogenous Mcl-1 in mouse embryonic fibroblasts (Fig. 1B). The BH3 region of Noxa alone seemed critical because marked degradation of Mcl-1 was also seen when this was placed in the context of the Bim_S backbone. However, a Bim BH3 domain did not reduce Mcl-1 levels, either in its native form or when placed in the context of Noxa (Fig. 1C). Instead, Bim BH3 binding promoted Mcl-1 stabilization, akin to that previously noted with Puma (24). Levels of another prosurvival protein, Bcl-2, remained constant regardless of whether Bim_s or Noxa were expressed.

Interestingly, mutating the unique lysine in the BH3 region of Noxa to a glutamic acid (see Noxa mt3 in Fig. 1A) did not affect its capacity to trigger Mcl-1 degradation (Fig. 1C). Furthermore, an inactive form of hNoxa (3E), unable to bind Mcl-1 (18), did not affect Mcl-1 levels. Likewise, expression of mBad, which is also unable to bind Mcl-1 (18), does not affect Mcl-1 levels. Strikingly, a chimeric mBad containing the hNoxa BH3 did not trigger Mcl-1 degradation, suggesting that regions outside of the BH3 may influence Noxa-induced degradation (Fig. 1C). Unlike hNoxa, mNoxa contains two BH3 domains. Both of these sequences induced Mcl-1 degradation when expressed in the context of the human Noxa backbone, albeit not as efficiently as the single human Noxa BH3 (Fig. 1D). This may reflect the lower binding affinities of the mouse Noxa BH3s, compared with the human one, for Mcl-1 (18).

Thus, Bim and Noxa BH3 domains may induce distinct conformational changes in Mcl-1, one stabilizing it and the other prompting its destruction. To test this, we determined and analyzed the three-dimensional structures of Mcl-1 bound to Bim BH3 and to Noxa BH3.

A Mouse/Human Mcl-1 Chimera That Retains a Human BH3-Binding Groove.

Previously, mouse Mcl-1 has been expressed in Escherichia coli with an N-terminal truncation of 151 residues (ΔN151) and a C-terminal truncation of 23 residues (ΔC23) to remove the transmembrane anchor (5). This leaves the BLR of the molecule in a form that can be expressed as a soluble protein (hereafter referred to as mMcl-1^BLR). However, the comparable human Mcl-1 construct (ΔN170 and ΔC21) did not produce soluble recombinant protein (data not shown). We have overcome this by constructing a chimera of mouse and human Mcl-1^BLR. This protein is identical to the human sequence with the exception of nine N-terminal amino acids [hereafter referred to as hMcl-1^BLR; see supporting information (SI) Fig. 5]. In particular, the region of the protein encompassing the BH3-binding groove is of human origin. Additionally, this protein can be expressed and purified in similar yields to mMcl-1^BLR. Isothermal titration calorimetry (ITC) confirms that hMcl-1^BLR and mMcl-1^BLR have similar binding characteristics for peptides derived from the BH3 regions of a variety of BH3-only proteins (SI Table 2 and data not shown).

The Structure of the hMcl-1^BLR:hBim BH3 Complex.

The refined model (Table 1; and see SI Table 3) includes residues D172 to S193 and G203 to E322 on hMcl-1^BLR and residues R53 to R75 on the helical Bim BH3 peptide. The protein adopts the canonical Bcl-2 family fold of eight α-helices. No interpretable electron density is evident for residues 194–202 in the linker region between helices α1 and α2, nor for residues 51, 52, and 76 at the end of the Bim peptide. The structure of the complex (Fig. 2 A and B) reveals the peptide bound to the homologous regions of Mcl-1 as compared with the previously reported structure of a complex between mouse Bcl-x_L and mouse Bim BH3 (Fig. 2C) (8). One strikingly conserved structural feature, not only with the Bcl-x_L:Bim complex but also with the CED-9:EGL-1 complex of Caenorhabditis elegans proteins (25), is the hydrogen bonding interaction between D67 on the peptide and a conserved arginyl residue in the BH1 domain of the antiapoptotic molecule (R263 in hMcl-1^BLR). This contact is displayed in SI Fig. 6, together with the experimentally phased electron density.

The surface of hMcl-1^BLR to which the peptide binds is a groove embellished by pockets of varying sizes and depth (Fig. 2B). The deepest of these accommodates L62 of the BH3 peptide and includes two water molecules. It is contiguous with the shallower pockets for I58 and I65. A saddle point in the groove aligns with G66 of the peptide, separating the pockets accommodating C-terminal residues in the peptide from the three pockets described above. Residues F69 and Y73 are both more solvent-exposed than any of the three N-terminal pocket-binding residues.

The Bim BH3 peptide is completely located within the binding groove. Because of the proximity of M231 of Mcl-1 to Bim residues I58 and L62, we measured the binding of the hBim BH3 to hMcl-1^BLR both in its native (methionyl) and derivatized (selenomethionyl) forms by ITC (SI Table 2). The dissociation constants observed for both are indistinguishable.

The electrostatic potential at the solvent-exposed surface of hMcl-1^BLR, calculated with the Bim BH3 peptide removed, is displayed in Fig. 2B. The BH3-binding groove of hMcl-1 is more positively charged than the corresponding region on mBcl-x_L (Fig. 2C) as described earlier for the mouse protein (5). The surface profiles of Mcl-1 and Bcl-x_L are also distinctly different, especially around the pockets for the first and fourth hydrophobic residues of the BH3 ligand.

The Crystal Structure of the hMcl-1^BLR:mNoxaB BH3 Complex.

Thus far, attempts to crystallize hMcl-1^BLR with a mNoxaA or a hNoxa BH3 peptide have not been successful. Here we present the crystal structure of hMcl-1^BLR in complex with the mNoxaB BH3 peptide. The model for this complex includes residues D172 to V321 of hMcl-1^BLR and residues D73 to N93 of mNoxaB. The crystals and resulting model are of lesser quality than those of the Bim complex (Table 1). The α1–α2 of hMcl-1^BLR linker is visible because of crystal contacts involving this segment. The mNoxaB BH3 peptide binds to hMcl-1^BLR in a similar fashion to the Bim BH3 peptide, but with one notable distinction. The N-terminal portion of the peptide, before Q77, is not helical (Fig. 3A) and is not intimately engaged with the binding groove of hMcl-1^BLR. Sequence alignments of BH3 domains (see Fig. 1A) indicate that the mNoxaB sequence is unique because it contains no hydrophobic residue at position 74. No electron density is observed for the N-terminal peptide sequence from residue 68 to residue 72, whereas the segment from 73 to 76 is neither helical nor intimately bound to hMcl-1^BLR. No ordered structure is evident for the carboxylate moieties of the peptide at residue D73 or E74 or for the peptide side chains of L87 and R88. An overlay of Mcl-1 from the Bim BH3 and mNoxaB BH3 complexes is displayed in Fig. 3B (see Discussion).

During the refinement of this structure it became apparent that cysteine residues on hMcl-1^BLR (C286) and the peptide (C75) of a crystallographically (6₃) related complex were close enough to form a disulfide bond. SDS/PAGE analysis of dissolved crystals (data not shown) confirmed the presence of this covalent linkage.

Solution Structure of the mMcl-1^BLR:mNoxaB BH3 Complex.

The NMR structure of the mMcl-1^BLR:mNoxaB BH3 complex (SI Table 4) was solved to determine the state of the mNoxaB BH3 N terminus in solution (Fig. 3C). Mouse Mcl-1^BLR was used for this purpose because the solution properties of the hMcl-1^BLR precluded a high-resolution structural study by NMR. The solution structure shows the mNoxaB peptide as helical from residue 72 to residue 93 and demonstrates that the N-terminal region of the peptide does indeed engage the Mcl-1 groove. This is the only significant difference observed between the solution (Fig. 3C) and crystal (Fig. 3A) structures. Amino acid sequence differences between human and mouse Mcl-1 (SI Fig. 5) reveal only one substitution in the binding groove, L246F, and that cannot account for this difference. Thus, the absence of helicity at the N terminus of the mNoxaB peptide described in the hMcl-1^BLR:mNoxaB crystal structure is a crystallization artifact. Mouse NoxaB residue E74 is tolerated at the conserved hydrophobic position because the charged carboxyl group is coordinated by K215 of mMcl-1 (K234 in the hMcl-1^BLR protein) at the solvent-exposed end of the peptide–protein interface (Fig. 3D).

Contribution of the N Terminus of mNoxaB to Mcl-1 Binding.

ITC was performed to confirm the role of the N terminus of mNoxaB in binding to the hMcl-1^BLR chimera. Removal of seven amino acids from the N terminus of the sequence together with the substitution C75A resulted in no reduction in the affinity (SI Table 2). However, removal of one more residue caused a 35-fold reduction in the binding constant. A similar truncation of the hNoxa BH3 resulted in no detectable binding of the peptide to hMcl-1^BLR.

These data suggest that the N-terminal region of mNoxaB BH3 may readily disengage from the hMcl-1^BLR binding groove, providing a rationale for the crystallization artifact involving the disulfide bond between neighboring protein–peptide complexes.

The Structural Determinant of BH3 Ligand-Induced Mcl-1 Degradation.

BH3 chimeras (Fig. 4A), with regions of Noxa BH3 replaced with those of Bim (which on its own did not cause Mcl-1 degradation) (Fig. 1C), point to a region toward the C terminus of the BH3 region being required for Mcl-1 degradation (Fig. 4B). The Noxa BH3 sequence (-FRQKLL-) appears to be essential because replacing this with corresponding Bim sequences (-AYYARR-, hNoxa/Bim 3) abrogated the effect on Mcl-1 stability. This chimeric BH3 peptide still binds Mcl-1 with low nanomolar affinity (SI Table 2).

Comparison of Mcl-1 in Complex with the BH3 Domains of Bim or Noxa.

The solution (Fig. 3C) and crystal structures (Fig. 3A) of the mNoxaB BH3 complexes are not significantly different, apart from at the N terminus of the peptide as described above. Binding studies of N-terminally truncated mNoxaB BH3 peptides suggest that this region does not make a major contribution to the binding energy with the Mcl-1 groove and that the structure seen in the crystalline state has arisen through capture of an unbound form of the N-terminal segment through a disulfide bond with a crystallographically neighboring hMcl-1^BLR:mNoxaB BH3 complex.

An overlay of the x-ray structures of Mcl-1 from the complexes with both Bim and mNoxaB is illustrated in Fig. 3B. Different crystal contacts are the likely explanation for minor differences in the two Mcl-1 structures at the α2–α3, α3–α4, and α5–α6 corners. Bearing in mind the medium resolution of the mNoxaB complex, no evidence is found of significant structural differences in the binding groove of hMcl-1^BLR reflecting adaptation to the different peptide ligands. However, the side chain orientations of Mcl-1 residues L235 and I237 are different in the two complexes, presumably because of the failure of the bound mNoxaB peptide to extend in a canonical helical conformation back toward its N terminus.

Because no significant differences are observed in the structures of Mcl-1 in the two peptide complexes, we conclude that it is the structure of the Mcl-1:Noxa complex itself that triggers its elimination rather than a Noxa-induced conformational change in Mcl-1.

The Structural Correlate of Degradation.

Steady-state levels of Mcl-1 in healthy cells are regulated in part by the E3 ubiquitin ligase Mule (14). Mule contains a BH3 sequence that binds to Mcl-1, but not to Bcl-x_L or Bcl-2, resulting in Mule-mediated Mcl-1 degradation. Here, we show that the Bim BH3 sequence stabilizes Mcl-1 levels. In a recent study, it was shown that the BH3-only protein Puma also stabilized Mcl-1 levels (24). One model by which Bim and Puma may stabilize Mcl-1 holds that their binding to Mcl-1 precludes that of Mule. Conversely, the Noxa BH3 region causes proteasomal-dependent Mcl-1 degradation (15), suggesting that this interaction may promote binding of an E3 ligase or some other adapter molecule, which is required for degradation to occur. Alternatively, the complex may be directly recognized by the proteasome itself. The structures of the Bim and Noxa BH3 peptides with Mcl-1 may reveal distinctive features that delineate the recognition epitope.

The sequences of BH3 domains of Bim and hNoxa differ markedly at both the N and C termini (Fig. 1A) but are similar throughout the central region of the domain. The properties of hNoxa mutant 3 (Fig. 1C) eliminate a role for K35 in degradation, although that lysine is a unique feature of Noxa among all of the mammalian BH3-only proteins. Mouse Noxa also causes Mcl-1 degradation (data not shown), but it has two BH3 motifs, and, at least in the context of human Noxa, both harbor this activity, although not as strongly as the hNoxa BH3. Within the hNoxa BH3, the sequence -FRQKLL- is important for degradation because its substitution by the sequence -AYYARR- from Bim leads to Mcl-1 stabilization, as seen for Bim itself. In the structure we report for the mNoxaB complex, the sequence is -LRQKLL-, and the glutaminyl residue projects toward the Mcl-1 surface with other amino acids in the sequence remaining highly solvent-accessible. One similarity among hNoxa, mNoxaB, and mNoxaA within this region is the presence of two hydrophobic residues, either -LL- or -AP-, at the end of the sequence. In contrast, the two BH3 domains that have been shown to stabilize Mcl-1 levels, Bim and Puma, have argininyl residues at both of these positions. Further experiments are needed to clarify the precise role of these residues.

Interestingly, unlike hNoxa BH3 in the context of the BH3-only protein Bim, hNoxa BH3 in the context of the BH3-only protein Bad does not promote destruction of Mcl-1. This result suggests that sequences outside of the BH3 domain of the protein Bad inhibit Mcl-1 degradation. Formally identifying such sequences may prove difficult because Bad (like Bim) is an intrinsically unstructured protein that assumes some structure in its BH3 domain only on engagement with Bcl-2 family proteins (26).

Collectively, these results support a model whereby the complex formed between Noxa and Mcl-1 is recognized for destruction, potentially by an E3 ligase or some other adapter molecule. Furthermore, the region of the complex recognized includes the C-terminal residues of the Noxa BH3 domain. Whether it also includes regions of the N-terminal PEST domain of Mcl-1, which is predicted to be unstructured, awaits clarification.

Comparison of Bound and Free Mcl-1.

The structures reported here invite comparisons with Bcl-x_L, with respect to the unliganded-to-liganded transition and to the liganded states with a common peptide ligand Bim BH3. Compared with the solution structure of mMcl-1^BLR with no peptide ligand (5), both of the complexes reported here have a more open binding groove (SI Fig. 7). The structural change that accompanies peptide binding is largest in the region of the central two hydrophobic pockets, h2 and h3. The Cα atoms of hMcl-1^BLR residues N223 and D256 are 6 Å further apart in the Bim BH3 complex than in the solution structure of the unliganded mouse molecule. The binding sites for the C terminus (beyond the conserved -GD- sequence) of both the Bim and mNoxaB BH3 peptides are preformed in the unliganded mMcl-1 structure. As predicted (5), unliganded Mcl-1 undergoes a smaller conformational change in binding a BH3 peptide than do Bcl-x_L (8) and CED-9 (25, 27).

Unique Features of the Mcl-1 Binding Groove.

The binding affinities of Mcl-1 and Bcl-x_L for certain BH3 domains vary over four or more orders of magnitude (18). In particular, the BH3 domain of Bad has no measurable binding to Mcl-1 whereas it binds tightly (nanomolar) to Bcl-x_L, Bcl-2, and Bcl-w. Interestingly, the recently described Bcl-x_L antagonist ABT-737 (19) mimics this binding profile. The high-resolution structure we describe here for hMcl-1 complexed with the Bim BH3 peptide reveals significant differences from the complex formed by Bcl-x_L (8).

The BH3-binding grooves of Mcl-1 and Bcl-x_L share many features that derive from common sequences within their BH1, BH2, and BH3 domains. However, three points of difference are noteworthy. (i) A single amino acid inserted into the BH1 domain of Mcl-1 is located in the α4–α5 corner. Two Mcl-1 residues (G257 and V258) occupy the space of a single residue (glycine) in Bcl-x_L (SI Fig. 8). As a consequence, R263 is somewhat less solvent-exposed in the Mcl-1:Bim BH3 complex than its homolog in the Bcl-x_L complex. It is this arginyl residue that participates in a conserved hydrogen-bonding interaction with an aspartyl residue of the peptide (D67 in this case) in all crystal structures of complexes reported to date. (ii) Compared with Bcl-x_L (and with Bcl-2 and Bcl-w), the BH3 domain sequence of Mcl-1 has distinctive differences adjacent to the conserved glycine and aspartic acid residues where the Mcl-1 sequence reads -VGDGVXXN- compared with -AGDEFXXR-. The consequence of these substitutions is to radically remodel the binding site for the C terminus of the peptide ligand, especially for the fourth hydrophobic residue (h4 in Fig. 1A), F69 in the case of Bim. The pocket accommodating this residue is less well defined and more solvent-exposed than its counterpart in Bcl-x_L (see Fig. 2). (iii) Helix α3 is well formed in the Mcl-1 complex but poorly so in the Bcl-x_L complex (SI Fig. 9). One result of this difference in the paths of the protein backbones (Q229 to K234 of Mcl-1 compared with S106 to Q111 of mouse Bcl-x_L) is that the binding pockets for the first two hydrophobic residues of the Bim BH3 sequence (h1-I58 and h2-L62) are more constricted and less contiguous in the Mcl-1 complex.

The Bcl-x_L complex is of mouse proteins (Protein Data Bank ID code 1PQ1), so the Bim peptide sequences are not identical. The only significant differences occur in the side chain conformations of the first and third of the four canonical hydrophobic residues of the BH3 motif, I58 and I65 (SI Fig. 9). These differences appear to compensate for corresponding structural differences between hMcl-1^BLR and Bcl-x_L described above. It is interesting to note that the sequence of Bad BH3, which is selective for Bcl-x_L over Mcl-1, has Tyr and Met, respectively, at these two positions.

Do any of these structural differences correlate with the failure of the BH3 domain of Bad to bind to Mcl-1? An earlier study (5) of selected point mutants of Mcl-1 failed to rescue binding to the Bad BH3, but a substitution in that BH3 domain of Y105I did recover some binding to wild-type Mcl-1. The differences between Mcl-1 and Bcl-x_L in α3 described above may bear on this result and suggest that binding to Bad BH3 might be enabled, at least partially, by relieving crowding around the h1 pocket. Note, however, that this crowding is largely due to a difference in the main chain conformations of Mcl-1 and Bcl-x_L in this region, a difference unlikely to be fully remedied by a single amino acid sequence change.

No structure has yet been reported for ABT-737 bound to Bcl-x_L, so one cannot directly attribute the differences described above to the selectivity of ABT-737 for Bcl-x_L. Nevertheless, the mere size of ABT-737 (≈800 Da) suggests that it engages a significant portion of the binding groove, whereon any or all of these differences may come into play to prevent binding to Mcl-1.

Retroviral Expression of BH3-Only Proteins.

Retroviral expression constructs were made by using the pMIG vector (MSCV-IRES-GFP; GFP sequence is that of EGFP) as described previously (18, 28). These plasmids were transiently transfected, by using Lipofectamine (Invitrogen, Carlsbad, CA), into Phoenix Ecotropic packaging cells (29). Filtered virus-containing supernatants were used to infect Bax^−/−Bak^−/− mouse embryonic fibroblasts by spin inoculation as described previously (15). Stable cell lines expressing vector alone, Bim_S, or Noxa variants were generated by selection of GFP⁺ mouse embryonic fibroblasts after retrovirus spin inoculation. After lysis of 1.5 × 10⁶ cells in buffer containing 1% (vol/vol) Triton X-100, proteins were resolved by SDS/PAGE and immunoblotted by using antibodies against Mcl-1, Bcl-2, and HSP 70 [N6; gift of W. Welch (University of California, San Francisco, CA) and R. Anderson (Peter MacCallum Cancer Centre, Melbourne, Australia)] as described (15). Fig. 1A describes the regions of sequence that were shuttled to construct hybrid BH3-only proteins. In the case of mBad, the chimeric BH3 domain replaced residues 97–122.

Protein Expression and Purification.

The hMcl-1^BLR chimera consists of residues 152–189 from mouse Mcl-1 and 209–327 from human Mcl-1 and was constructed by using the conserved SphI site. Mouse Mcl-1^BLR (amino acids 152–308) and hMcl-1^BLR (171–327) were expressed in E. coli BL21 (DE3) as GST fusion proteins as previously described (5). Cells were grown in superbroth at 37°C and induced with 1 mM isopropyl β-d-thiogalactoside (IPTG) at an OD₆₀₀ of 0.3. After 3 h, cells were harvested and lysed in buffer 1 (50 mM Tris, pH 8.0/150 mM NaCl/1 mM EDTA). Supernatant was applied to a glutathione Sepharose column (GE Healthcare, Piscataway, NJ) and then washed with buffer 1. On-column cleavage was performed with Prescission Protease (GE Healthcare). Mcl-1 was eluted with buffer 1 and further purified on a Superdex 200 column (GE Healthcare) in 50 mM Tris (pH 8.0)/150 mM NaCl. Selenomethionine-labeled protein was expressed as described (30) and purified as per native protein.

Mouse NoxaB BH3 peptide for use in NMR was prepared as previously outlined (26). Mass spectrometry was used to confirm peptide homogeneity. Isotopically labeled proteins were grown in minimal media by using ¹⁵NH₄Cl and d-[U-¹³C]glucose as the sole nitrogen and carbon sources, respectively, as described (31). Mouse Mcl-1/NoxaB complex was prepared by titration by using NMR to detect the end point. Two samples were prepared, ¹³C,¹⁵N-labeled Mcl-1/unlabeled NoxaB-BH3 and unlabeled Mcl-1/¹³C,¹⁵N-labeled NoxaB-BH3. NMR samples contained ≈0.5 mM protein in 50 mM sodium phosphate (pH 6.7), 70 mM NaCl, and 0.04% sodium azide in H₂O:²H₂O (95:5). Synthetic peptides were purchased from Mimotopes (Victoria, Australia).

Crystallography, Data Collection, and Structure Determination.

For the hMcl-1^BLR:hBim BH3 complex, selenomethionine-labeled protein was mixed with an equimolar amount of peptide and concentrated to 12 mg/ml. Crystals of the complex were grown in hanging drops at 22°C [reservoir solution: 0.2 M ZnCl₂/0.2 M imidazole, pH 5.75/2 mM tris(2-carboxyethyl)phosphine (TCEP)]. Before flash-freezing in liquid N₂, crystals were equilibrated into cryoprotectant consisting of reservoir solution and increasing concentrations of trehalose (final trehalose concentration, 30%). X-ray data were collected at three wavelengths on beamline X29A at the National Synchrotron Light Source (Brookhaven National Laboratory). Data were integrated and scaled with HKL2000 (32). Two Se sites were found by using data from all three wavelengths using HKL2MAP (33). An initial model was built from the resulting map using COOT (34). Several rounds of building and refinement in REFMAC5 (35) led to the final model shown in Table 1.

For the hMcl-1^BLR:mNoxaB BH3 complex, native protein was mixed with an equimolar amount of peptide and concentrated to 15 mg/ml. Crystals were grown in hanging drops at 22°C (reservoir solution: 12% PEG 4000/4% isopropanol/5% dioxane/0.1 M Tris, pH 8.0). Crystals were equilibrated into cryoprotectant consisting of reservoir solution and increasing concentrations of ethylene glycol (final concentration, 25%) before flash-freezing. A native x-ray data set was collected on beamline X-29A at the National Synchrotron Light Source. Data were integrated and scaled with HKL2000 (32) (Table 1). The structure was determined by molecular replacement with PHASER (36–38) using the coordinates of hMcl-1^BLR from the hMcl-1^BLR:hBim BH3 complex as a search model. Several rounds of building in COOT and refinement with REFMAC5 led to the model described in Table 1.

NMR Spectroscopy.

Spectra were recorded at 25°C on a Bruker DRX 600 600-MHz spectrometer equipped with triple-resonance probes and pulsed-field gradients or AV-500 and AV-800 spectrometers equipped with a cryogenically cooled probes, operating at 500 and 800 MHz, respectively. A series of heteronuclear 3D NMR experiments were recorded by using either ¹⁵N or ¹³C,¹⁵N double-labeled mMcl-1^BLR (39). Spectra were processed by using TOPSPIN (Bruker, Billerica, MA) and analyzed by using XEASY (40).

Distance restraints were measured from the 120-ms mixing time 3D ¹⁵N-edited NOESY, ¹³C-edited NOESY, and 2D NOESY spectra. Hydrogen bond constraints were applied within α-helices at a late stage of the structure calculation (4). φ and ψ backbone torsion angles were derived by using TALOS (41). Dihedral angle restraints for φ and ψ angles were used as summarized in SI Table 4. ³J_HNHα were derived from a 3D HNHA spectrum (42).

Initial structures were calculated by using CYANA 2.1 (43) and optimized to obtain low target functions and refined with X-PLOR-NIH 2.14 (44) by using the OPLSX nonbonded parameters in explicit water (45). Structural statistics for the final set of 20 structures, chosen on the basis of their stereochemical energies, are presented in SI Table 4. PROCHECK_NMR (46) and MOLMOL (47) were used for the analysis of structure quality. The final structures had no experimental distance violations >0.3 Å or dihedral angle violations >5°. Structural figures were generated in MOLMOL.

ITC.

ITC was performed by using a VP-ITC microcalorimeter (Microcal, Amherst, MA). Experiments were performed in 20 mM Tris (pH 8.0)/150 mM NaCl at 25°C. Titrations consisted of 42 7-μl injections of peptide at 40 μM into 1.34 ml of protein at 5 μM. Data were analyzed by using Origin software (OriginLab, Northampton, MA).